Long-term time series forecasting has extensive applications across various critical domains. However, its modeling process encounters significant challenges, including difficulties in capturing long-range dependencies and issues arising from distribution shifts. As the prediction horizon extends, time series exhibit complex multiscale dynamic characteristics. Meanwhile, data non-stationarity causes trained models to struggle in adapting to future distributional changes, thereby substantially affecting forecasting performance. To alleviate the aforementioned challenges, this paper proposes an attention-based forecasting model against distribution shift (AFMDS). The proposed method aims to improve prediction accuracy and robustness under non-stationary environments. Specifically, this method employs a global-local decomposition framework to effectively separate shared long-term dynamics across variables from individual short-term fluctuations, thereby enhancing the model’s capability to capture both long-range dependencies and inter-variable correlations. To mitigate the negative effects of non-stationarity, a lightweight statistical normalization module reconstructs time series data based on core statistical characteristics. Additionally, a novel non-stationary attention mechanism is designed to dynamically restore essential temporal information. Experimental results on benchmark datasets demonstrate that this model outperforms existing SOAT method, showcasing superior forecasting accuracy and robustness in scenarios involving long-term prediction and distribution shifts.

With the rapid growth of data, feature selection has become essential for improving machine learning performance. However, most existing unsupervised feature selection methods rely on greedy strategies, which often lead to suboptimal solutions. Moreover, traditional information–theoretic approaches are primarily designed for discrete data and require discretization when applied to continuous data, potentially causing information loss. To address these issues, this paper proposes a global unsupervised feature selection method based on fuzzy mutual information (UFS-FMI). The proposed method integrates fuzzy set theory with information measures to quantify feature relevance and redundancy, and formulates a fractional optimization model. A combination of projection neural networks and kWTA neural networks is employed to achieve global optimization. Experimental results on nine UCI benchmark datasets demonstrate that UFS-FMI consistently outperforms several representative methods in terms of classification accuracy, clustering accuracy, and normalized mutual information (NMI). In particular, on datasets such as Movement_libras, Ionosphere, and Control, the proposed method achieves significantly improved classification performance, confirming its effectiveness and robustness.

In this paper, we consider the workflow scheduling problem with soft deadlines and fuzzy time uncertainties in cloud manufacturing environments. Workflow tasks in cloud manufacturing often involve uncertain execution and logistics times due to large-scale and geographically distributed resources, creating significant challenges for efficient and reliable scheduling. To address these challenges, we propose the Tolerance-aware Neighborhood Search (TANS) algorithm, which integrates fuzzy time quantization with heuristic neighborhood search techniques. A comprehensive workflow scheduling architecture is established, and multiple neighborhood structures and heuristic search methods are developed to systematically explore feasible solutions. The effectiveness of TANS is verified by extensive experiments and parameter calibrations based on Analysis of Variance (ANOVA). Experimental results indicate that TANS reduces workflow delays by 39% on average compared to state-of-the-art methods, demonstrating high efficiency in scenarios with different numbers of tasks and resources.